Normalization is essential but must be selectively applied based on
the unique characteristics of each dataset and the specific biological
questions at hand.
Techniques like SCTransform and log scaling offer ways to balance
technical correction with biological integrity.
Examining both raw and normalized data can provide comprehensive
insights into the absolute and relative characteristics of cellular
components in spatial transcriptomics.
Feature selection is a crucial step in spatial transcriptomics
analysis, particularly for non-variance-stabilizing normalization
methods like NormalizeData.
Techniques such as VST and mean-variance plotting enable researchers
to focus on genes that provide the most biological insight.
Different proportions of highly variable genes and feature selection
methods can significantly influence the analytical outcomes, emphasizing
the need for tailored approaches based on the specific characteristics
of each dataset.
Linear dimensionality reduction methods like PCA are crucial for
initial data simplification and noise reduction.
Nonlinear methods like UMAP are valuable for detailed exploration of
data structures post-linear preprocessing.
The sequential application of PCA and UMAP can provide a
comprehensive view of the spatial transcriptomics data, leveraging the
strengths of both linear and nonlinear approaches.
Differential expression testing pinpoints genes with significant
expression variations across regions or clusters.
Moran’s I statistic reveals spatial autocorrelation in gene
expression, critical for examining spatially dependent biological
activities.
Moran’s I algorithm effectively identifies genes expressed in
anatomically distinct regions, as validated from the correlation
analysis with the DE genes from the annotated regions.